1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
|
# Copyright 2016 Google Inc. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Library for getting system information during TensorFlow tests."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import ctypes as ct
import platform
import tensorflow as tf
from tensorflow.core.util import test_log_pb2
def _gather_gpu_devices_proc():
"""Try to gather NVidia GPU device information via /proc/driver."""
dev_info = []
for f in tf.gfile.Glob("/proc/driver/nvidia/gpus/*/information"):
bus_id = f.split("/")[5]
key_values = dict(
line.rstrip().replace("\t", "").split(":", 1)
for line in tf.gfile.GFile(f, "r"))
key_values = dict(
(k.lower(), v.strip(" ").rstrip(" "))
for (k, v) in key_values.items())
info = test_log_pb2.GPUInfo()
info.model = key_values.get("model", "Unknown")
info.uuid = key_values.get("gpu uuid", "Unknown")
info.bus_id = bus_id
dev_info.append(info)
return dev_info
class CUDADeviceProperties(ct.Structure):
# See $CUDA_HOME/include/cuda_runtime_api.h for the definition of
# the cudaDeviceProp struct.
_fields_ = [
("name", ct.c_char * 256),
("totalGlobalMem", ct.c_size_t),
("sharedMemPerBlock", ct.c_size_t),
("regsPerBlock", ct.c_int),
("warpSize", ct.c_int),
("memPitch", ct.c_size_t),
("maxThreadsPerBlock", ct.c_int),
("maxThreadsDim", ct.c_int * 3),
("maxGridSize", ct.c_int * 3),
("clockRate", ct.c_int),
("totalConstMem", ct.c_size_t),
("major", ct.c_int),
("minor", ct.c_int),
("textureAlignment", ct.c_size_t),
("texturePitchAlignment", ct.c_size_t),
("deviceOverlap", ct.c_int),
("multiProcessorCount", ct.c_int),
("kernelExecTimeoutEnabled", ct.c_int),
("integrated", ct.c_int),
("canMapHostMemory", ct.c_int),
("computeMode", ct.c_int),
("maxTexture1D", ct.c_int),
("maxTexture1DMipmap", ct.c_int),
("maxTexture1DLinear", ct.c_int),
("maxTexture2D", ct.c_int * 2),
("maxTexture2DMipmap", ct.c_int * 2),
("maxTexture2DLinear", ct.c_int * 3),
("maxTexture2DGather", ct.c_int * 2),
("maxTexture3D", ct.c_int * 3),
("maxTexture3DAlt", ct.c_int * 3),
("maxTextureCubemap", ct.c_int),
("maxTexture1DLayered", ct.c_int * 2),
("maxTexture2DLayered", ct.c_int * 3),
("maxTextureCubemapLayered", ct.c_int * 2),
("maxSurface1D", ct.c_int),
("maxSurface2D", ct.c_int * 2),
("maxSurface3D", ct.c_int * 3),
("maxSurface1DLayered", ct.c_int * 2),
("maxSurface2DLayered", ct.c_int * 3),
("maxSurfaceCubemap", ct.c_int),
("maxSurfaceCubemapLayered", ct.c_int * 2),
("surfaceAlignment", ct.c_size_t),
("concurrentKernels", ct.c_int),
("ECCEnabled", ct.c_int),
("pciBusID", ct.c_int),
("pciDeviceID", ct.c_int),
("pciDomainID", ct.c_int),
("tccDriver", ct.c_int),
("asyncEngineCount", ct.c_int),
("unifiedAddressing", ct.c_int),
("memoryClockRate", ct.c_int),
("memoryBusWidth", ct.c_int),
("l2CacheSize", ct.c_int),
("maxThreadsPerMultiProcessor", ct.c_int),
("streamPrioritiesSupported", ct.c_int),
("globalL1CacheSupported", ct.c_int),
("localL1CacheSupported", ct.c_int),
("sharedMemPerMultiprocessor", ct.c_size_t),
("regsPerMultiprocessor", ct.c_int),
("managedMemSupported", ct.c_int),
("isMultiGpuBoard", ct.c_int),
("multiGpuBoardGroupID", ct.c_int),
# Pad with extra space to avoid dereference crashes if future
# versions of CUDA extend the size of this struct.
("__future_buffer", ct.c_char * 4096)]
def _gather_gpu_devices_cudart():
"""Try to gather NVidia GPU device information via libcudart."""
dev_info = []
system = platform.system()
if system == "Linux":
libcudart = ct.cdll.LoadLibrary("libcudart.so")
elif system == "Darwin":
libcudart = ct.cdll.LoadLibrary("libcudart.dylib")
elif system == "Windows":
libcudart = ct.windll.LoadLibrary("libcudart.dll")
else:
raise NotImplementedError("Cannot identify system.")
version = ct.c_int()
rc = libcudart.cudaRuntimeGetVersion(ct.byref(version))
if rc != 0:
raise ValueError("Could not get version")
if version.value < 6050:
raise NotImplementedError("CUDA version must be between >= 6.5")
device_count = ct.c_int()
libcudart.cudaGetDeviceCount(ct.byref(device_count))
for i in range(device_count.value):
properties = CUDADeviceProperties()
rc = libcudart.cudaGetDeviceProperties(ct.byref(properties), i)
if rc != 0:
raise ValueError("Could not get device properties")
pci_bus_id = " " * 13
rc = libcudart.cudaDeviceGetPCIBusId(ct.c_char_p(pci_bus_id), 13, i)
if rc != 0:
raise ValueError("Could not get device PCI bus id")
info = test_log_pb2.GPUInfo() # No UUID available
info.model = properties.name
info.bus_id = pci_bus_id
dev_info.append(info)
del properties
return dev_info
def gather_gpu_devices():
"""Gather gpu device info.
Returns:
A list of test_log_pb2.GPUInfo messages.
"""
try:
# Prefer using /proc if possible, it provides the UUID.
dev_info = _gather_gpu_devices_proc()
if not dev_info:
raise ValueError("No devices found")
return dev_info
except (IOError, ValueError):
pass
try:
# Fall back on using libcudart
return _gather_gpu_devices_cudart()
except (OSError, ValueError, NotImplementedError):
return []
|
